Re: 2.17.2 regtests

[address@hidden]

On 13 sept. 2012, at 10:17, David Kastrup <address@hidden> wrote:

> "address@hidden" <address@hidden> writes:
> 
>> On 13 sept. 2012, at 08:31, David Kastrup <address@hidden> wrote:
>> 
>>> "address@hidden" <address@hidden> writes:
>>> 
>>>> Just a note that Les Nereides will go back to its old state w/ the
>>>> revert of the fix for 2801.  I still need help debugging it,
>>>> especially w/ advice about profiling tools.
>>> 
>>> Correct me if I am wrong, but if you have a phrasing slur across 20
>>> note columns, doesn't that mean that you get 20 phrasing slur stubs that
>>> all want to be scored?
>> 
>> Stubs are created for every note column but most of them are suicided
>> so that you only get one stub per vertical axis group.
> 
> Well, either they are used for scoring/collision avoidance before you
> start the suicide session or not.  If they are, you will pay that cost.

They are not.

> If they are not, it is utterly pointless to create them before you
> actually know how many of them you will actually need.

In LilyPond engravers, grobs should be announced at the timestep during which 
their left-bounding column is created.  The few exceptions to this 
(Auto_beam_engraver is the only one I can think of) cause of bugs (almost all 
the bugs in the tracker, for example, that say "auto beams don't do X" are 
because of this).  In general, when an engraver contains acknowledge_X, it 
acknowledges grob X under the assumption that grob X was created at that 
timestep.  Ditto for acknowledge_end_X.

So, it is better to announce a grob and then kill it if you don't need it (this 
is done with Flags and Ties, for example - see Stem_engraver and Tie_engraver 
respectively) rather than announce it after its timestep.

> 
>> In this case, there are two vertical axis groups, so two stubs.
> 
> Still 20 get created and managed, right?
> 

Yup.

Cheers,
MS